Refactor custom_preprocess() for more robustness

CONTRIBUTING.md

@@ -139,7 +139,7 @@ The following methods are expected to be implemented in the metric class:
 - `evaluate_instance()`: Gets model and data for a single instance as input, returns evaluation result.
 
 The following methods are optimal for implementation:
-- `custom_preprocess()`: In case `general_preprocess()` from base class is not sufficient, additional preprocessing steps can be added here.
+- `custom_preprocess()`: In case `general_preprocess()` from base class is not sufficient, additional preprocessing steps can be added here. This method must return a dictionary with string keys or None. If a dictionary is returned, additional keyword arguments can be used in `evaluate_instance()`.
 - `custom_postprocess()`: Additional postprocessing steps can be added here that is added on top of the resuling evaluation scores.
 
 ### Using Helpers

quantus/metrics/axiomatic/completeness.py

@@ -255,8 +255,6 @@ def evaluate_instance(
         y: np.ndarray,
         a: np.ndarray,
         s: np.ndarray,
-        c: Any,
-        p: Any,
     ) -> bool:
         """
         Evaluate instance gets model and data for a single instance as input and returns the evaluation result.

quantus/metrics/axiomatic/input_invariance.py

@@ -237,8 +237,6 @@ def evaluate_instance(
         y: np.ndarray,
         a: np.ndarray,
         s: np.ndarray,
-        c: Any,
-        p: Any,
     ) -> bool:
         """
         Evaluate instance gets model and data for a single instance as input and returns the evaluation result.
@@ -257,10 +255,6 @@ def evaluate_instance(
             The explanation to be evaluated on an instance-basis.
         s: np.ndarray
             The segmentation to be evaluated on an instance-basis.
-        c: any
-            The custom input to be evaluated on an instance-basis.
-        p: any
-            The custom preprocess input to be evaluated on an instance-basis.
 
         Returns
         -------
@@ -298,9 +292,7 @@ def custom_preprocess(
         a_batch: Optional[np.ndarray],
         s_batch: np.ndarray,
         custom_batch: Optional[np.ndarray],
-    ) -> Tuple[
-        ModelInterface, np.ndarray, np.ndarray, np.ndarray, np.ndarray, Any, Any
-    ]:
+    ) -> None:
         """
         Implementation of custom_preprocess_batch.
 
@@ -325,19 +317,6 @@ def custom_preprocess(
             In addition to the x_batch, y_batch, a_batch, s_batch and custom_batch,
             returning a custom preprocess batch (custom_preprocess_batch).
         """
-
-        custom_preprocess_batch = [None for _ in range(len(x_batch))]
-
         # Additional explain_func assert, as the one in prepare() won't be
         # executed when a_batch != None.
         asserts.assert_explain_func(explain_func=self.explain_func)
-
-        return (
-            model,
-            x_batch,
-            y_batch,
-            a_batch,
-            s_batch,
-            custom_batch,
-            custom_preprocess_batch,
-        )

quantus/metrics/axiomatic/non_sensitivity.py

@@ -259,8 +259,6 @@ def evaluate_instance(
         y: np.ndarray,
         a: np.ndarray,
         s: np.ndarray,
-        c: Any,
-        p: Any,
     ) -> int:
         """
         Evaluate instance gets model and data for a single instance as input and returns the evaluation result.
@@ -279,10 +277,6 @@ def evaluate_instance(
             The explanation to be evaluated on an instance-basis.
         s: np.ndarray
             The segmentation to be evaluated on an instance-basis.
-        c: any
-            The custom input to be evaluated on an instance-basis.
-        p: any
-            The custom preprocess input to be evaluated on an instance-basis.
 
         Returns
         -------
@@ -330,9 +324,7 @@ def custom_preprocess(
         a_batch: Optional[np.ndarray],
         s_batch: np.ndarray,
         custom_batch: Optional[np.ndarray],
-    ) -> Tuple[
-        ModelInterface, np.ndarray, np.ndarray, np.ndarray, np.ndarray, Any, Any
-    ]:
+    ) -> None:
         """
         Implementation of custom_preprocess_batch.
 
@@ -349,29 +341,16 @@ def custom_preprocess(
         s_batch: np.ndarray, optional
             A np.ndarray which contains segmentation masks that matches the input.
         custom_batch: any
-            Gives flexibility ot the user to use for evaluation, can hold any variable.
+            Gives flexibility to the user to use for evaluation, can hold any variable.
 
         Returns
         -------
         tuple
             In addition to the x_batch, y_batch, a_batch, s_batch and custom_batch,
             returning a custom preprocess batch (custom_preprocess_batch).
         """
-
-        custom_preprocess_batch = [None for _ in range(len(x_batch))]
-
         # Asserts.
         asserts.assert_features_in_step(
             features_in_step=self.features_in_step,
             input_shape=x_batch.shape[2:],
         )
-
-        return (
-            model,
-            x_batch,
-            y_batch,
-            a_batch,
-            s_batch,
-            custom_batch,
-            custom_preprocess_batch,
-        )